Skip to content

REGISTRY.md

Latest commit

 

History

History
52 lines (36 loc) · 3.15 KB

REGISTRY.md

File metadata and controls

52 lines (36 loc) · 3.15 KB
Raw

🏅 SkillSpec Conformance Registry

A public list of repositories whose skills conform to SkillSpec. Two sides of the same standard live in this folder:

  • README.md certifies agents — does your agent use skills correctly (discover, restrain, follow structure, self-verify, refuse fabrication)?
  • This file certifies skill repositories — do the skills themselves reach a SkillSpec conformance level (L1 Loadable → L2 Structured → L3 Trustworthy) with a clean security scan?

The point is adoption: SkillSpec is only a standard if repos other than this one can measure themselves against it and prove it. The validator is published (skillspec-check on npm) and has no dependencies, so anyone can.

Earn the badge (any repo, 30 seconds)

npx skillspec-check skills/ --min-level 3     # gate your CI on L3 (exit 1 if any skill is below)
npx skillspec-check skills/ --badge > badge.json   # emit a shields.io endpoint object

The --badge output is a shields.io endpoint object whose message is the minimum level across all your skills (a catalog is only as trustworthy as its weakest skill) and the count. Commit badge.json and point a shields endpoint at its raw URL:

![SkillSpec](https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/<you>/<repo>/main/badge.json)

The badge is earned, not decorative — it recomputes from your actual skills every CI run, so it can't drift from reality.

Get listed here

  1. Make your repo pass at the level you're claiming: npx skillspec-check <dir> --min-level <1|2|3>.
  2. PR one entry to registry.json: 
    { "name": "<repo>", "url": "https://github.com/<you>/<repo>", "skills": <n>, "level": <1|2|3>, "verifiedAt": "<YYYY-MM-DD>" }
  3. CI validates your entry's shape (node conformance/verify-registry.mjs), and a maintainer runs the one-line validator against your repo — npx skillspec-check <your repo> --min-level <level> — to confirm the claimed level before merge. No self-attestation: the level is reproducible by anyone, including you, from the published tool.

Conformance levels

Level Name Bar
L1 Loadable Valid frontmatter with a name and a decision-useful description (an agent can tell when to fire it).
L2 Structured Declared inputs and a concrete output template — two runs produce the same shape of artifact.
L3 Trustworthy L2 plus self-verification: Quality Checks and Anti-Patterns. The skill knows when its own output is wrong.

All levels also pass the security scan (no instruction-override, exfiltration, or embedded-credential patterns).

Registry

Repo Skills Level Verified
pm-claude-skills 750 L3 2026-07-21
(be next — PR above)

Honest scope: SkillSpec conformance is structural + security, not a quality score. An L3 skill is well-formed and self-checking; whether it's good is what the eval --badge (measured/unmeasured) on each skill answers separately.